[Perf][CUDA] Programmatic dependent launch for the DSA decode kernels (#50230)

xiaozhoupy committed 3 months ago

[Model] Fix weight prefix mapping for native Qwen3.5 text-only checkp… (#50355)

zofia committed 3 months ago

[HPC Attention Backend] hpc attention backend support bf16 kv cache with fp8 weight (#50980)

Cheng Jiang committed 3 months ago

[ROCm] Fix AITER all-reduce fusion coverage (#50802)

Andreas Karatzas committed 3 months ago

[BugFix][KV Cache] Fix hybrid prefix caching with hidden-state extraction (#51108)

Canlin Guo committed 3 months ago

[Linear] [Kernel] add block-wise scaled_mm (#49932)

zofia committed 3 months ago

[XPU] update warning of XPU Graph (#50236)

liuzhenwei committed 3 months ago

[Bugfix][Quantization] Fix MXFP4 conversion for FlashInfer CUTLASS (#51038)

Gabriel Wu committed 3 months ago

[Bugfix][Spec Decode] Auto-enable async scheduling for draft models (#48341)

JooHo Lee committed 3 months ago

[XPU] Route AWQ linear through choose_mp_linear_kernel (#50840)

zofia committed 3 months ago

[XPU][Test] Support MultiConnector accuracy testing on XPU (#51160)

liuzhenwei committed 3 months ago

[XPU] Support MXFP8 linear weights for INC DeepSeek V4 model (#48476)

Xiaochang Wu committed 3 months ago

[MM][CG] Support ViT full CUDA graph for Ernie-4.5-VL image inference (#45254)

Qiuyang Yue committed 3 months ago

[BugFix][Pooling] Skip weight-prefix probe when model has WeightsMapper (#50890)

Li Juqi committed 3 months ago

[Bugfix][MM] Fix MiniCPM-V placeholder replacement and image processor loading on Transformers v5 (#48413)

Yunzhu Lu committed 3 months ago

[Feature] Parse request priority from HTTP header (#51089)

Chauncey committed 3 months ago

[Docs] List Intel XPU attention backends (#51215)

baodi committed 3 months ago

[Hardware] Use torch.accelerator.empty_host_cache() for host cache cl… (#51107)

liuzhenwei committed 3 months ago

[Model Runner V2] Cache draft logits in model's LM head dtype (#50910)

Giancarlo Delfin committed 3 months ago

K3: remove the add operation for megamoe path (#51146)

Jee Jee Li committed 3 months ago